Deep learning model detects and classifies femoral neck fractures with 93% accuracy in retrospective study

A retrospective multicenter diagnostic study evaluated a deep learning-based automatic detection and classification model for femoral neck fractures using hip imaging. The study included 10,010 hip images from 806 patients across four Chinese hospitals, with the model compared against 12 physicians of varying experience levels. The model achieved a five-fold cross-validation mean accuracy of 93.34% and specificity of 95.29%, with an independent test set mean AUC of 95.78%. When physicians used the model, resident physicians' diagnostic accuracy reportedly improved markedly, and the diagnostic gap between resident and senior clinicians narrowed.

Safety and tolerability data were not reported in this diagnostic accuracy study. The retrospective design represents a key limitation, as the authors note prospective randomized studies are needed to confirm clinical utility. The study was conducted at four Chinese hospitals, which may limit generalizability to other settings or populations.

The findings show promise for this deep learning model as a clinical decision support tool for femoral neck fracture detection and Garden classification. However, clinicians should interpret these results cautiously as evidence of association only from a retrospective diagnostic study. The practice relevance remains preliminary until prospective validation confirms whether the model improves diagnostic outcomes in real-world clinical practice.